8652a685e6
- rewrite README with project purpose, standard workflow, corrected index - add docs/NEW_PLATFORM_GUIDE.md (new GPU onboarding SOP, GLM5.2 reuse) - fix ../../scripts/common -> ../../../scripts/common in 42 experiment scripts - refresh stale docs (EXPERIMENT_GUIDE, H200_QUICKSTART, ADAPTIVE_CONCURRENCY_USAGE, BENCHMARK_WORKFLOW) - remove dead code (dp_proxy.py) and .bak leftovers - add p800 adaptive results (tp4_dp2/tp8_dp1 metrics + summary) - gitignore envs/charts and .tmp_charts
DSV4 H200 vLLM Baseline Benchmark
NVIDIA H200 + native vllm + DeepSeek-V4-Flash baseline benchmark experiment.
This experiment uses:
/data/user1/yy/envs/vllmas the server environment/data/user1/yy/envs/sglangas the benchmark client environment
Quick Start
# Run the full experiment (start server + benchmark + parse)
bash experiments/dsv4_h200_vllm/run_bench.sh
# Reuse an already-running server
SKIP_MANAGE_SERVER=1 bash experiments/dsv4_h200_vllm/run_bench.sh
Results land in experiments/dsv4_h200_vllm/results/<RUN_ID>/.
Configuration
Edit config.env or override via environment variables:
MODEL_PATH=/data/models/DeepSeek-V4-Flash \
PORT=30005 \
SCENARIOS="32 512 256 128 512 256" \
bash experiments/dsv4_h200_vllm/run_bench.sh
Files
| File | Purpose |
|---|---|
config.env |
Experiment-level configuration (model, port, venv paths, scenarios) |
start_server.sh |
Start a plain vLLM baseline server for DeepSeek-V4-Flash |
run_bench.sh |
Orchestrator: metadata → server → benchmark → parse |
parse_results.py |
Parse raw JSONL outputs into results.json + report.md |
Platform
This experiment targets platforms/nvidia_h200.env.